Research On Communication Signal Modulation Recognition Algorithm Based On Time Frequency Image

In the field of the signal processing,the algorithm of signal modulation recognition has always been the important task.In the more and more complex electronic environment,the researchers are trying their best to deal with the communication signals,the study of this part can be used in the reality of the war environment,and it has the most important significance for the complex environment.By dealing with the problems and disadvantages for this stage algorithms,in this paper in order to solve the problems that they were less consider about the connections between different characters of the signals,the convolutional neural network will be used in this paper.By using the smooth pseudo-Wigner distribution time-frequency analysis to make the one-dimensional signals change into the color time-frequency images,besides that in order to improve the performance of the final result,the texture features were extracted for the feature fusion operation.The results of the simulation show that if compared with the other algorithms which were not consider the connections between different characters,the algorithm of this paper can improve the result in some way at the low SNR,the content of this paper shown in these points:1.By using the time-frequency transformation,change the IQ signal which format are one-dimensional into the color time-frequency images,use the digital image processing technological.2.In this paper extracted six types of features from the IQ signals by using the traditional signal feature extraction algorithm and using the five types of classifiers to classify the digital signals.3.Make the TF-CNN(Time Frequency Convolutional Neural Networks)architecture,use it to extract features from the time-frequency images,fusion this type of features with the LBP texture feature,then reduction the features through the PCA dimensionality algorithm.4.Build the TF-Resnet(Time-Frequency Resnet)architecture,make the time-frequency images to be the input for the network to get the signals' time-frequency features,after making the signals' LBP feature and CNN's feature fusion together,input them into the GA-KELM classifier for signal classification.The results of the simulation experiments show that the TF-CNN and TF-Resnet of this paper have the superior classification performance at the low signal to noise ratio,they can achieve the high precision signal automatic modulation recognition and classification.